aboutsummaryrefslogtreecommitdiffstats
path: root/test/test_download.py
diff options
context:
space:
mode:
authorPhilipp Hagemeister <phihag@phihag.de>2014-08-21 11:52:07 +0200
committerPhilipp Hagemeister <phihag@phihag.de>2014-08-21 11:52:07 +0200
commite8ee972c6ea211a51275c8eb8bfc81b9ae3a9844 (patch)
tree4ab2920fb57fc777c28af30cf4c38166c81005d2 /test/test_download.py
parentaf8322d2f993653dbec5e6ad943f8cc632b901e4 (diff)
downloadyoutube-dl-e8ee972c6ea211a51275c8eb8bfc81b9ae3a9844.zip
youtube-dl-e8ee972c6ea211a51275c8eb8bfc81b9ae3a9844.tar.gz
youtube-dl-e8ee972c6ea211a51275c8eb8bfc81b9ae3a9844.tar.bz2
Allow playlist test definitions in test_download.
This moves playlist tests where they belong, i.e. to the extractors themselves. Additionally, all our network interaction configuration for tests in test_download now applies to playlist tests as well.
Diffstat (limited to 'test/test_download.py')
-rw-r--r--test/test_download.py32
1 files changed, 27 insertions, 5 deletions
diff --git a/test/test_download.py b/test/test_download.py
index d654058..c8d4ec2 100644
--- a/test/test_download.py
+++ b/test/test_download.py
@@ -63,15 +63,21 @@ def generator(test_case):
def test_template(self):
ie = youtube_dl.extractor.get_info_extractor(test_case['name'])
other_ies = [get_info_extractor(ie_key) for ie_key in test_case.get('add_ie', [])]
+ is_playlist = any(k.startswith('playlist') for k in test_case)
+ test_cases = test_case.get(
+ 'playlist', [] if is_playlist else [test_case])
+
def print_skipping(reason):
print('Skipping %s: %s' % (test_case['name'], reason))
if not ie.working():
print_skipping('IE marked as not _WORKING')
return
- if 'playlist' not in test_case:
- info_dict = test_case.get('info_dict', {})
- if not test_case.get('file') and not (info_dict.get('id') and info_dict.get('ext')):
+
+ for tc in test_cases:
+ info_dict = tc.get('info_dict', {})
+ if not tc.get('file') and not (info_dict.get('id') and info_dict.get('ext')):
raise Exception('Test definition incorrect. The output file cannot be known. Are both \'id\' and \'ext\' keys present?')
+
if 'skip' in test_case:
print_skipping(test_case['skip'])
return
@@ -81,6 +87,9 @@ def generator(test_case):
return
params = get_params(test_case.get('params', {}))
+ if is_playlist and 'playlist' not in test_case:
+ params.setdefault('extract_flat', True)
+ params.setdefault('skip_download', True)
ydl = YoutubeDL(params)
ydl.add_default_info_extractors()
@@ -93,7 +102,6 @@ def generator(test_case):
def get_tc_filename(tc):
return tc.get('file') or ydl.prepare_filename(tc.get('info_dict', {}))
- test_cases = test_case.get('playlist', [test_case])
def try_rm_tcs_files():
for tc in test_cases:
tc_filename = get_tc_filename(tc)
@@ -105,7 +113,10 @@ def generator(test_case):
try_num = 1
while True:
try:
- ydl.download([test_case['url']])
+ # We're not using .download here sine that is just a shim
+ # for outside error handling, and returns the exit code
+ # instead of the result dict.
+ res_dict = ydl.extract_info(test_case['url'])
except (DownloadError, ExtractorError) as err:
# Check if the exception is not a network related one
if not err.exc_info[0] in (compat_urllib_error.URLError, socket.timeout, UnavailableVideoError, compat_http_client.BadStatusLine) or (err.exc_info[0] == compat_HTTPError and err.exc_info[1].code == 503):
@@ -121,6 +132,17 @@ def generator(test_case):
else:
break
+ if is_playlist:
+ self.assertEqual(res_dict['_type'], 'playlist')
+ expect_info_dict(self, test_case.get('info_dict', {}), res_dict)
+ if 'playlist_mincount' in test_case:
+ self.assertGreaterEqual(
+ len(res_dict['entries']),
+ test_case['playlist_mincount'],
+ 'Expected at least %d in playlist %s, but got only %d' % (
+ test_case['playlist_mincount'], test_case['url'],
+ len(res_dict['entries'])))
+
for tc in test_cases:
tc_filename = get_tc_filename(tc)
if not test_case.get('params', {}).get('skip_download', False):